floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Refs #174. Fixed the overflowing buffer bug of multithreading hbmv and sbmv. 

Instead of using thread 0 buffer, each thread uses its own sb buffer.
Thus, it can avoid overflowing thread 0 buffer.
This commit is contained in:
Zhang Xianyi 2013-02-13 16:05:58 +08:00
parent 5c8bf6ae0e
commit 5155e3f509
2 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
driver/level2/sbmv_thread.c
View File

@ -65,7 +65,6 @@ static int sbmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
a = (FLOAT *)args -> a; a = (FLOAT *)args -> a;
x = (FLOAT *)args -> b; x = (FLOAT *)args -> b;
y = (FLOAT *)args -> c;
lda = args -> lda; lda = args -> lda;
incx = args -> ldb; incx = args -> ldb;
@ -76,6 +75,10 @@ static int sbmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
n_from = 0; n_from = 0;
n_to = n; n_to = n;
//Use y as each thread's n* COMPSIZE elements in sb buffer
y = buffer;
buffer += ((COMPSIZE * n + 1023) & ~1023);
if (range_m) { if (range_m) {
n_from = *(range_m + 0); n_from = *(range_m + 0);
n_to = *(range_m + 1); n_to = *(range_m + 1);
@ -83,7 +86,6 @@ static int sbmv_kernel(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, F
a += n_from * lda * COMPSIZE; a += n_from * lda * COMPSIZE;
} }
if (range_n) y += *range_n * COMPSIZE;
if (incx != 1) { if (incx != 1) {
COPY_K(n, x, incx, buffer, 1); COPY_K(n, x, incx, buffer, 1);
@ -331,7 +333,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
if (num_cpu) { if (num_cpu) {
queue[0].sa = NULL; queue[0].sa = NULL;
queue[0].sb = buffer + num_cpu * (((n + 255) & ~255) + 16) * COMPSIZE; queue[0].sb = buffer;
queue[num_cpu - 1].next = NULL; queue[num_cpu - 1].next = NULL;
exec_blas(num_cpu, queue); exec_blas(num_cpu, queue);
@ -344,7 +346,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
#else #else
ONE, ZERO, ONE, ZERO,
#endif #endif
buffer + range_n[i] * COMPSIZE, 1, buffer, 1, NULL, 0); (FLOAT*)(queue[i].sb), 1, buffer, 1, NULL, 0);
} }
AXPYU_K(n, 0, 0, AXPYU_K(n, 0, 0,

1
driver/others/blas_server.c
View File

@ -385,6 +385,7 @@ static int blas_thread_server(void *arg){
+ GEMM_ALIGN) & ~GEMM_ALIGN)) + GEMM_OFFSET_B); + GEMM_ALIGN) & ~GEMM_ALIGN)) + GEMM_OFFSET_B);
} }
} }
queue->sb=sb;
} }
#ifdef MONITOR #ifdef MONITOR